Wildfire server clustering / scaling out


there is a lot of discussion about how to enhance the scalability of wildfire installations. it is obvious that doing non-blocking i/o on the wildfire server (ws) itself and the connection managers (cm) will improve overall performance significantly. however this still leaves us with a single wildfire server which will be a bottleneck. adding a servernode2servernode ‘‘cluster-protocol’’ (which is a subject in the pampero roadmap) will solve this.

however, >currently< the best scale out solution we have is a bunch of cm’'s and ONE ws.

do you guys see an easy way to create a ‘‘pseudo’’ wildfire server cluster like this:

  • all data is stored in a central db server / db server cluster

  • there are several ws which have cm’'s in front of them, they use the standard xmpp server2server protocol to perform the message routing

what would be the limitiations here (would the jid’'s look consistent, etc.)?



Without some sort of clustering feature, it wouldn’'t be possible to have multiple core Wildfire servers all acting as the same domain. We certainly plan to support clustering in the future, however.